Thermal expansion is inevitable for space structures under the alternating temperature of outer space around the earth. This may lead to the thermal stress and deformation due to the mismatch of the coefficient of thermal expansion. Near-zero thermal expansion (Near-ZTE) is a vitally essential demand for large-scale space telescopes or antennas to preserve their spatial precision and resolution. Recently, mechanical metamaterials with superior and tailorable properties have attracted significant interest with regard to developing negative materials or ultra-property materials. In this paper, the near-ZTE space structure architected by a dual-hourglass bi-material lattice is achieved by the structural optimization method with the gradient-based algorithm. First, an hourglass lattice with adjustable structural parameters is optimized to seek the design of effective negative thermal expansion (NTE) in the thickness direction. Then, two building blocks with both NTE and legacy positive thermal expansion (PTE) are combined as a dual-layered lattice to obtain the near-ZTE. Finally, a structure with near-ZTE of about ~10⁻⁹ m/(m·K) is obtained. Furthermore, the various lattice configurations, such as the hexagonal pyramid and triangle pyramid, are investigated in detail. Finally, the natural frequencies of two near-ZTE lattices are calculated by the modal analysis method, and the stiffness is discussed for the optimal solution of space applications. This work demonstrates that the near-ZTE structure can be achieved by utilizing the negative metamaterial and structural optimization method. It provides a novel solution to design the large-scale space structures with the near-zero thermal induced deformation, and may be constructed and assembled by the on-orbit fabrication technology.

Open-hole features are widely used in composite structures but inevitably alter intrinsic load-bearing behavior and fracture characteristics. This study develops an artificial intelligence-based method for surface damage morphology prediction of multi-hole carbon fiber reinforcement thermoplastic laminates. A pix2pix conditional generative adversarial network incorporating a rule-based ideal fracture mask is established to predict geometry-dependent fracture morphologies and evaluate its predictive capability for unseen geometric configurations. Four global pixel-based and five morphology-specific metrics are employed to assess the agreement between predicted and experimental damage morphologies. The results show that the proposed network successfully reproduces the dominant crack paths, critical-hole locations, and distributed surface-splitting patterns observed experimentally. Ablation analysis further confirms that ideal fracture mask provides effective guidance for fracture morphology prediction. Despite the limited training dataset, the model retains reliable predictive capability for unseen regions of the geometric parameter space and accurately identifies critical failure holes. Quantitative comparison of single-hole laminates further shows similar fracture characteristics under quasi-static and fatigue loading, indicating potential extension to fatigue applications. However, broader applicability requires validation using datasets with multi-hole fatigue specimens. Overall, the proposed method provides an image-based framework for predicting surface damage morphology under quasi-static loading and localizing fracture-critical regions within the predicted morphology.